DSL has become the technology of choice for delivery of high bandwidth data over copper access links, replacing legacy TDM (Time Division Multiplexing) services. This is due to the required telephony access infrastructure being almost universally present, and to the continuous increase in DSL bit-rates.
We can distinguish two types of services delivered by DSL:
1. Synchronous services, such as legacy TDM (E1s or T1s), which require distribution of the associated service clock.
2. Asynchronous data services, such as Internet or asynchronous voice applications (e.g., cellular compressed voice calls).
Synchronous services were the first deployed, with DSL lines replacing TDM links. For example, HDSL (ITU-T Recommendation G.991.1) and later SHDSL (ITU-T Recommendation G.991.2) were designed to replace standard T1 (1.544 Mbit/s) or E1 (2.048 Mbit/s) services, and extended the applicability and range of these services. However, the focus rapidly shifted to asynchronous broadband data services. For example, ADSL (Asymmetric DSL) (and all its derivatives) and VDSL (Very High Speed DSL) are commonly used to extend an asynchronous packet network (e.g., Ethernet, IP, or MPLS) to customer sites.
Due to the initial focus on synchronous services, DSL standards include built-in mechanisms to distribute the service clock, referred to as Network Timing Reference or NTR. However, current Digital Subscriber Line Access Multiplexers (DSLAMs) are optimized for asynchronous data services, and are thus often not equipped with NTR functionality. This does not impact their main aim of providing asynchronous services to residential customers, but hinders the provision of synchronous services, such as TDM pseudowires.
For synchronous services the service clock needs to be accurately delivered to the end-application in order to prevent buffer overflow and bit errors. Moreover, even if the service itself is fully asynchronous, the end equipment might still need a good reference clock for its operation. For example, cellular base-stations require a very accurate and stable clock to derive their RF transmission frequency. In the past, such base-stations derived their clock from the incoming TDM link, but with the replacement of TDM links by DSL lines optimized for asynchronous services, this inherent frequency distribution is lost.
FIGS. 1, 2 and 3 depict various prior art clock distribution schemes for DSL technologies. FIG. 1 shows a Network Timing Reference (NTR) clock distribution scheme, such as described in U.S. Pat. No. 6,937,613 to Bedrosian. As can be seen in FIG. 1, at the Central Office (CO) side, the Local Timing Reference (LTR) 30 within the central office DSL Access Multiplexer (DSLAM) 17, is locked to an external timing source 1. Such a timing source is typically a Primary Reference Clock (PRC) with long-term frequency accuracy within ±10−11 of Coordinated Universal Time (UTC). Since the DSL mapper 18, which maps the incoming payload 31 into the transmitted sequence of symbols 19, is fed with the local timing reference, the physical-layer of the transmission is locked to the external timing source. At the remote side, the DSL de-mapper 20 within the Customer Premises Equipment (CPE) 13 recovers the external timing source information from the incoming symbols rate and uses this recovered clock information 15 to transmit the recovered payload 14 to the end-user equipment 16.
A similar approach is used in ITU-T Recommendation G991.2 (SHDSL) where the DSLAM's clock may be locked to a Network Timing Reference (NTR), which is an 8 KHz clock typically traceable to a PRC. The remote modem may then extract NTR timing information from the physical layer. While ITU Recommendation G991.2 does not mandate this NTR functionality, most SHDSL DSLAMs do support it.
FIG. 2 depicts a clock distribution scheme that uses the SRTS (Synchronous Residual Time Stamp) method, as described in Reissued Pat. No. 36,633 to Fleischer et al. This scheme enables remote recovery of a reference clock without locking the physical-layer clock. Instead, the difference between the reference clock and the physical-layer clock is encoded and transmitted, enabling the remote terminal to mathematically reconstruct the desired reference clock frequency. This principle has been extended to general networks, and is now called ‘common clock frequency recovery’ (see ITU-T Recommendation Y.1413 subclause 10.1.3) or ‘differential clock recovery’ (see ITU-T Recommendation G.8261 subclause 8.2).
As illustrated in FIG. 2, SRTS encoder 21 within clock distribution 3 has access to both the reference clock that needs to be distributed and network clock 39 (i.e., the common clock). The momentary phase difference between these two clocks is encoded into four bit SRTS words that are periodically delivered to packet assembly 23 (via interface 22) where they are assembled into packets and sent across packet network 4. At the other end of the network in clock recovery 37, the four bit SRTS words are extracted in packet disassembly 38 and periodically delivered to the differential clock recovery 27 (via interface 15) that also has access to the same network clock 39. Using this network clock and the four bit SRTS words, the original service clock can be recovered on output line 29.
The SRTS method generally provides a very accurate recovered clock as it is not affected by impairments introduced by higher network layers such as Packet Delay Variation (PDV). Nevertheless, a basic requirement for its use is CO and customer premises access to a common clock. Having such a common clock is, in many networks and applications, not possible for various reasons; hence, for such cases other clock recovery techniques must be used. ITU-T Recommendation G.992.1 (ADSL), and its later variants ITU-T Recommendations G.992.3, G.992.4, and G.992.5, support an indirect NTR mechanism, based on similar principles. Rather than directly locking the physical-layer symbol clock to the external frequency reference, the physical-layer clock of ADSL and its derivatives is locked to a Local Timing Reference (LTR), and the phase difference between the external reference and LTR is periodically transmitted. The DSLAM encodes this phase difference in four bits, and places these in a fixed location within the ADSL frame. At the CPE these four bits are extracted from the ADSL frame and are used in combination with the recovered physical-layer clock (LTR) to re-generate the original NTR clock. VDSL (ITU-T Recommendation G.993.1) uses the same method of NTR distribution, however it encodes the phase difference between NTR and LTR in eight bytes.
Similar to the situation for SHDSL, the ADSL and VDSL standards do not mandate NTR support. However, unlike the situation described above, where the majority of SHDSL DSLAMs support NTR in practice, ADSL and VDSL DSLAMs most often do not support it. As aforementioned, this is due to ADSL and VDSL being primarily used to provide asynchronous services to end users, where NTR transport functionality is not required.
The lack of NTR transport support within the DSLAM necessitates the deployment of other frequency distribution means for those end applications that require an accurate frequency reference. U.S. Pat. No. 5,440,313 to Osterdock et al. describes the use of a GPS receiver as a frequency reference. GPS receivers can indeed provide a stable and accurate frequency reference to end applications, however they suffer the drawbacks of being relatively expensive, involving costly and complicated installation procedures, and are only applicable where GPS can be reliably received (e.g., where roof-top access is possible). In similar fashion, a dedicated TDM link can sometimes be provided purely for frequency distribution.
FIG. 3 presents the general concept of Adaptive Clock Recovery (ACR) as applied to DSL systems without NTR support. This concept is based on the more general principle of adaptive clock recovery based solely on the arriving data packet flow, such as taught by U.S. Pat. No. 5,396,492 to Lien. That invention includes a destination node that receives data packets from a network and stores them in a buffer. The data packets are read out of the buffer using a locally generated clock. The fill level of the buffer is monitored and used to control the frequency of the locally generated clock, thus adapting the locally generated clock to the reference clock at the source of the packet flow.
Dedicated timing packets are sent from a master clock distribution 3 located somewhere within the core network. This clock distribution unit receives a clock reference traceable to a Primary Reference Clock (PRC) and periodically transmits dedicated timing packets conveying frequency information, to all CPEs 13. Such dedicated timing packets could belong to a constant rate TDM pseudowire flow or could be time distribution protocol packets, e.g., according to IEEE (Institute of Electrical and Electronic Engineers) standard 1588-2008 or to the IETF (Internet Engineering Task Force) Network Time Protocol (NTP) described in RFC (Request For Comments) 1305. These timing packets traverse the packet network 4 and are directed by DSLAM 10 to the relevant CPE 13. Arriving at CPE 13 these packets are used by the ACR function 33 to regenerate a frequency reference locked to the source PRC, resulting in the End User Equipment (EUE) 16 receiving frequency traceable to a PRC via clock interface 29.
The scheme of FIG. 3 suffers from a substantial drawback. All ACR methods rely directly or indirectly on the arrival times of timing packets. Packet arrival times are distorted by Packet Delay Variation (PDV) introduced by the end-to-end path. DSL links in particular tend to introduce considerable PDV, precluding ACR-based schemes from conformance to the stringent frequency distribution standards required by some applications.
There is therefore a need for inexpensively providing an accurate and reliable substitute for NTR for DSLAMs that do not support standard NTR.